The long-term goals of my research are (1) to understand how cortical networks flexibly compute actions from sensory and cognitive signals and (2) to use this knowledge to develop brain computer interfaces (BCIs) that restore the ability to perform useful actions to humans with paralysis. The specific objective of this project is to determine how interconnected networks of neurons in the premotor and motor cortex compute reach and grasp actions from different types of signals. Dorsal premotor (PMd) and ventral premotor (PMv) areas of frontal cortex appear to represent separate processing streams that converge on primary motor cortex (MI) where reach and grasp are unified. This interacting circuit appears to generate commands for actions by dynamically linking various cues to specific arm and hand actions, but studies conflict on the nature of the signals processed, the types of sensorimotor transforms performed, or their roles in arm and hand movement. The present study introduces several new approaches to understand how the collective dynamics of these cortical networks provide a flexible substrate to link different signals and actions. Chronically implanted multi-electrode arrays implanted in PMv, PMd and MI and full upper limb motion capture will be used to simultaneously record neuronal spiking (192 channels) and measure the actions of the arm, wrist and fingers while monkeys perform tasks in different contexts. Context will be varied by training monkeys to perform four tasks that differ in the signals used for to guide arm actions: (1) using ongoing sensory feedback to capture a swinging object, (2) planning different grasp types from visual cues, (3) planning different goals for reach (either point or grasp), (3) making a perceptual judgments to determine grasp-type. Context-dependent changes will be measured in single cell responses and population representation (SA1) and in network functional connectivity (SA2) in PMv, PMd and MI simultaneously recorded for each task. Recording two or more contexts within single sessions will allow direct comparison of the same ensembles to determine how context modifies coding in single neurons, how population models generalize across contexts, and how networks combine various signals to achieve coordinated reach and grasp. We will address the hypothesis that PMv and PMd form separate information processing channels by comparing neural coding features and context sensitivity between areas. Finally, SA3 will test the requirement for specific sensorimotor computations by disrupting processing in these networks using emerging, highly-selective optogenetic methods or electrical stimulation. These studies will reveal mechanisms by which cortical circuits perform sensorimotor transformations and flexibly link groups of neurons to generate voluntary behavior. In addition, they will help to determine whether signals available in premotor areas, as well as MI, can provide useful sources for BCI command signals. Such signals could allow people with paralysis to regain complex functions like reach and grasp through robotic limbs that could operate usefully across a range of contexts.